Development of a web interface for estimating melanoma from dermoscopic images using artificial intelligence techniques based on deep learning

Aim: Recently, hospitals and healthcare organizations increasingly need clinical decision support systems (CDSS) that can provide physicians and healthcare professionals with tailored patient assessments or recommendations. CDSS provides an important opportunity for sensitive medical applications, increases the work efficiency of the hospital and reduces costs. In this case, with the proposed thesis, it is aimed to create a model that can successfully predict melanoma using dermoscopic images and to develop a web-based system for classification of images with this model. Material and Method: In this study, a total of 24.268 images were used for model training and testing. 18.607 of the images contain benign lesions and 5661 melanomas. VGG16 pre-trained networks were used to build the model. Results: When the findings of our thesis study were examined, the developed KKDS was able to classify melanoma images with an accuracy of 84.94%. Other classification performance criteria were calculated as sensitivity value of 74.00% and selectivity value of 87.45%, respectively. In the analysis of the existing test images with the created model, 4000 healthy and 1000 cancerous images containing melanoma were used. The model predicted that of the 4000 images containing the images of healthy patients, 3498 were intact while they were intact, and 502 were melanoma even though they were intact. At the same time, the developed model classified 251 of 1000 images containing melanoma patient images as healthy while melanoma and 749 as melanoma while melanoma. Conclusion: Classification of melanoma, one of the deadliest skin cancers, has been carried out and made available to users via the web interface. This web interface enables physicians to make faster decisions for early diagnosis and diagnosis. On the other hand, the software is compatible with mobile as it uses Html5 infrastructure. In this way, it allows physicians to analyze with their mobile devices.